config.json

{
  "activation_dropout": 0.0,
  "activation_function": "relu",
  "angle_loss_coefficient": 1,
  "architectures": [
    "MonoDETRForMultiObjectDetection"
  ],
  "attention_dropout": 0.0,
  "auxiliary_loss": true,
  "backbone": "resnet50",
  "backbone_config": null,
  "bbox_cost": 5,
  "bbox_loss_coefficient": 5,
  "center3d_cost": 10,
  "center3d_loss_coefficient": 10,
  "class_cost": 2,
  "cls_loss_coefficient": 2,
  "d_model": 256,
  "decoder_attention_heads": 8,
  "decoder_depth_residual": true,
  "decoder_ffn_dim": 256,
  "decoder_layers": 3,
  "decoder_n_points": 4,
  "decoder_self_attn": true,
  "depth_loss_coefficient": 1,
  "depth_map_loss_coefficient": 1,
  "depth_max": 60.0,
  "depth_min": 0.001,
  "dim_loss_coefficient": 1,
  "disable_custom_kernels": false,
  "dropout": 0.1,
  "encoder_attention_heads": 8,
  "encoder_ffn_dim": 256,
  "encoder_layers": 3,
  "encoder_n_points": 4,
  "focal_alpha": 0.25,
  "freeze_backbone": true,
  "giou_cost": 2,
  "giou_loss_coefficient": 2,
  "group_num": 11,
  "id2label": {
    "0": "Pedestrian",
    "1": "Car",
    "2": "Cyclist"
  },
  "init_box": false,
  "init_std": 0.02,
  "init_xavier_std": 1.0,
  "is_encoder_decoder": true,
  "label2id": {
    "Car": 1,
    "Cyclist": 2,
    "Pedestrian": 0
  },
  "model_type": "mono3dvgv2",
  "num_channels": 3,
  "num_depth_bins": 80,
  "num_feature_levels": 4,
  "num_queries": 50,
  "position_embedding_type": "sine",
  "pretrained_backbone_path": "pretrained-models/resnet50/model.safetensors",
  "pretrained_dpt_path": "pretrained-models/depth-anything-v2-small",
  "region_loss_coefficient": 1,
  "torch_dtype": "float32",
  "transformers_version": "4.46.3",
  "two_stage": false,
  "use_pretrained_backbone": true,
  "use_timm_backbone": true,
  "use_torchvision_backbone": false,
  "with_box_refine": true
}